1 September 2026
In plain English
The machinery is back. For eight days every automated job in this system had stopped running. It restarted overnight; I dispatched three jobs myself this morning and all three worked. The daily brief can reach your inbox again, and the fact-checking and demand-sizing tools work.
On the forecasts: nothing moved, and one of them not moving is the point. Yesterday I worked out that I had been counting the same bad news about the telehealth company Hims & Hers over and over, lowering the odds each time. The same headlines came round again this morning. I checked the dates, found every underlying event about a month old, and left the number alone.
One genuinely new fact: US regulators approved the diabetes drug Mounjaro to also lower the risk of heart attack and stroke.
Signal detail
Today's digest resurfaced the Hims cluster — TipRanks "Visa, FTC Turn Up Heat", Yahoo "down 14.6% after raising guidance", both 2026-08-29 — and I nearly moved on it. One search settled it: the FTC complaint (with Utah and LA County, N.D. Cal.) was filed 2026-07-29, Visa's Acquirer Monitoring Program enrolment reported 2026-08-21, the securities solicitations 2026-08-24. All predate yesterday's upward correction, which existed because the 0.80→0.55 slide had read this same cluster as arriving evidence seven runs running. Share-price reaction to a month-old lawsuit is not new evidence. Historian, Case 2: the binding constraint on a compounded-peptide launch is the bulks-list rulemaking, not the news cycle, and that clock runs past this horizon.
No release from the US Medicare agency. Today's coverage cluster is state cost legislation and Medicare Advantage overpayment commentary, neither of which counts enrollees.
Nothing touched The Compound's own affiliate mix. Standing fragility unchanged: the Katalys payout denominator has been stale since 2026-08-24, so "unchanged" is partly "unmeasured" — and that is now fixable, because Actions is back. peptide-safety-incident: holds 0.95, at the ceiling, escalated again. Two more items today (Citeline on a peptide supply error, 2026-08-28; Stanford Report on compounded GLP-1 safety, 2026-08-29), neither from the five outlets the forecast names. See §3. Augur pick. A site called "Peptides TIL" launched by newswire on 2026-08-29 (OCNJ Daily, FinancialContent), covering stacks, where-to-buy, regulation and retatrutide data. That is bestpeptideforthat's lane, and the where-to-buy half is the monetisation our own site refuses on purpose. Promotional copy, so it is logged as an argument to watch rather than a fact — the PRESS-RELEASE lesson from the histogram, applied to myself the morning I read it.
Resolutions & upcoming
Nothing resolved. Nothing resolves in the next 14 days; the next horizon is 2026-12-31. Zero overdue.
`peptide-safety-incident` needs an operator decision, not another run. It has sat at the 0.95 ceiling for over a week, structurally unable to move, while its own criterion may already have been met by the July WaPo and CNN features. Ra may not resolve it; every further run just restates that.
How Barque got smarter today
- Order-defect channel: 124 defects across 35 orders — 116 corrected-and-shipped, 8 that stopped an order. Ranked: UNCLASSIFIED 31, OMISSION 18, PREMISE-FALSE 11, then STALE-EVIDENCE and INFERENCE at 7. With `(not enumerated) 25`, 56 of 124 — about 45% — is unreadable; this is the visible part, not the whole.
- Rank 1 (UNCLASSIFIED), mechanical change. Not a class, a hole: resolvers wrote `CLASS=PREMISE-INVERTED`, `CLASS=STRATA-MIX`, `SRCTYPE=ORDER`, `SRCTYPE=SECONDARY`, all outside the enums, so `validate` exits FAIL. Those are good names the taxonomy lacks. Filed as `barque-fix-defect-class-srctype-enum-gap`; that order is the change, not this bullet.
- Rank 2 (OMISSION), mechanical change. The pattern is an order asserting a commercial opportunity without checking the cheapest disconfirming fact — the Ro order argued a head term while Ro was absent from a 62-provider payout file. Committed `VERIFY` line: on my next order asserting an opportunity against a named provider, I check that provider against the Katalys payout file before the row is written, and state the result either way.
- Rank 3 (PREMISE-FALSE), brief amendment — verbatim, since I cannot edit my own brief. Add to STEP V as Tier 0b: "Before writing any row that asserts an opportunity or a gap, state the founding premise as one sentence beginning 'This order is only worth doing if…' and check that sentence against the cheapest available source. A premise that cannot be stated in one sentence is a theme, not a premise, and the order is not ready. PREMISE-FALSE and OMISSION are 29 of the 93 classified defects in the fortnight to 2026-09-01, more than any single verification class, and neither is caught by verifying DATED, NUMBERED or LEGAL specifics — in both, the specifics are usually right and the reason for acting is wrong."
- `SRCTYPE` is `NOT-RECORDED` on 29 of 35 rows, so for 83% the "which of my sourcing steps produced this" answer does not exist. Not inferring it. The missing field is its own finding.
- A near-miss on myself. My first openFDA pull showed a Mounjaro label with no cardiovascular indication and I nearly logged a contradiction against the trade press. The query lacked `sort=effective_time:desc`, and `meta.last_updated` describes the dataset, not the record. Re-queried sorted: same answer — but the primary confirms the approval, so the finding inverts to the label has not propagated, a caveat now written into the work order. Same shape as the PREMISE-INVERTED defect in this very digest.
- Reroutes: two rows, both already re-filed on earlier runs. No re-filing today, per the do-not-duplicate rule. Separately, `work-orders.tsv` lines 13–14 carry 10 columns against a 9-column schema — pre-existing from 2026-05-02, handed to the open `barque-fix-duplicate-work-order-id-guard` rather than padding the queue.